Relations between climate and plague. (a) Empirical distributions of the pixel-wise correlations between annual NDVI and the climate proxies. Only glacial accumulation is centered around zero for time lag 0, as accumulation rate the previous winter seems to be the relevant factor, while for T and S temporal autocorrelation gives both t = 0 and t = -1 significant relationships with NDVI. (b) Empirical distributions of the pixel-wise correlations between annual NDVI and sylvatic plague in the present (t = 0) and following (t = +1) year, both for observed (P) and estimated (Y) values. As for climate, plague seems non-randomly related to NDVI, both in the current and the preceding year. (c) GCV values for all 104 sylvatic plague models (eq. 8) vs. their correlation with human plague cases 1904-1948 (D). The 5% lowest-GCV models that form the estimate of climatically-forced sylvatic plague (Y) are highlighted in red. The blue dots show the mean D for each increment of GCV, connected with a (blue) trend line showing their close correlation (ρ = -0.99). (d) The empirical density distributions of D for the 5% (i.e., 500) best sylvatic plague models shown in red. They are very unlikely to be centered on zero, suggesting a nonrandom relationship between a climate-driven model's ability to predict sylvatic and human plague.